¹ I am glad to express here my thanks to Dr. Henry Hoenigswald and the members of the linguistic seminar at the University of Pennsylvania for valuable criticism and for linguistic material. I am particularly indebted to Dr. Roman Jakobson for an interesting conversation on the phonetic breakdown and grouping of phonemes. I owe an exceptionally heavy debt to Dr. Bernard Bloch, who has helped me state many of the more difficult points.

² This example of phonetic components is given here only for introductory simplicity. The analysis presented below is primarily distributional rather than phonetic.

^2a E.g. it is this technique that enables us, in languages which have a phonemic tone no each vowel (Fanti, Chinese, etc.), to extract the tones as separate phonemic elements.

^2b Voiced-voiceless sequences like /rp/ in carpet are not counted here, since /r/ has no voiceless homorganic counterpart.

³ The Prague Circle more closely approached the technique of dividing elements into simultaneous components, but purely on arbitrary phonetic grounds, when they said that the difference between two phonemes was not a vs. b, but a vs. a + x (where x is a Merkmal denoting the extra features which differentiate b from a). See N. S. Trubetzkoy, Grundzüge der Phonologie 67 (Travaux du Cercle Linguistique de Prague 7; 1939).

⁴ Y. R. Chao gives other ‘cases of one homogeneous sound represented by two or three piece symbols, each of which represented some aspect or aspects of the sound’ in his article The Non-Uniqueness of Phonemic Solutions of Phonetic Systems, Bulletin of the Institute of History and Philology 4.371 (Academia Sinica; Shanghai, 1934).

⁵ Z. S. Harris, The Phonemes of Moroccan Arabic, JAOS 62.309–18 (1942).

^5a It is possible, by Fourier analysis, to replace periodic waves by a sum of simpler periodic waves. The original waves (e.g. sound waves) can then be considered the resultants which are obtained by adding together all their component waves.

⁶ E.g. vibration of the vocal cords, giving ‘voice’. This might be the phonetic value of a particular component in a particular position, whereas the phonetic value of a particular phoneme in a particular position might be, for example, voice plus closing off of the nose plus closing of the lips (English /b/). A phonetic system of this kind without the phonemic limitation is Otto Jespersen's analphabetic system, presented in his Lehrbuch der Phonetik (2d ed. Leipzig and Berlin, 1912) and elsewhere.

⁷ Bernard Bloch, Phonemic Overlapping, American Speech 16.278–84 (1941).

^7a The lengths are not absolute (so many hundredths of a second) but relative. This means that an allophone [p], for instance, is not composed of two shorter allophones ‘p-closure’ and ‘p-release’.

^7b We will assume that these allophones satisfy all the criteria for phonemes—that is, that complementary allophones have been grouped together—except that allophones having different stress and pitch have been considered different sounds and hence not grouped together under one symbol. It is impossible to obtain the conventional phonemes until intonational components have been extracted from the allophones.

⁸ This operation is used implicitly throughout phonemics to keep us from breaking sounds down into smaller and smaller segments ad infinitum. We do not consider the lip closing and the lip opening of intervocalic /p/ to be separate phonemes, because they always occur together in that position.

⁹ A fuller discussion of the character of these contour components is given in fn. 22a below.

^9a Linguistic forms which are written in components will be set between diagonals, in the same way as forms written phonemically. It is convenient to use identical brackets for these two systems, because many linguistic forms cited in this paper are written partly in phonemes, partly in components: e.g. /†sz'r/ ‘tree’. We write in components only those parts of a form which are under discussion. This is permissible because phonemics is merely a special case of component analysis; the extension from phonemics into components can be carried out to any degree desired. In the analysis of Moroccan Arabic cited in fn. 5, the phonemes are of the usual kind except for the component /'/ (§2.3 above), which is included among the phonemes.—In some cases, where it is clear that a symbol indicates a component, the diagonals are omitted. The use of non-alphabetic marks like the horizontal bar (§5.2) is not in general desirable; but only such marks can depict on paper the effect of a long component that extends over more than one phoneme.

¹⁰ More briefly: Given that xy occurs, we select u such that xu does not occur. Then y = γ + u (where γ has 2-phoneme length, when two phonemes are present), and x = γ + w, where w is selected so that wu occurs.

¹¹ Note that in this example y does not occur alone.

¹² Aside from an unrelated limitation between /s/ and contiguous /z/.

¹³ Or of a word, except for one enclitic. That is, when † occurs, it extends from one word juncture to the next.

¹⁴ See fn. 12.

¹⁵ The morphophonemic alternation of φ for π + ʽ (e.g. in ἀφ' v) can also be avoided, if the ʽ component is written where it is heard. In the few cases of two aspirates within a word, a second ʽ would have to be written over the extra aspirate, and the statement of the length of ʽ would have to be adjusted accordingly. In the case of the reduplication there is a real elimination of a morphophonemic statement: the stem initial in /'pépūka/ (or /pép'ūka/) is /p/, which is duly present in the reduplication.

¹⁶ For the phonetic values of these phonemes see op.cit. in fn. 5.

¹⁷ The laryngal value for the ends of the labial component is not essential, though it seemed most convenient for various reasons. It would also be possible to assign merely a ‘non-labial’ value to the ends of the labial component, leaving it to the components in the neighboring positions to decide whether they are laryngal, dental, or palatal. They cannot be labial because a ‘non-labial’ component extends over them.

¹⁸ Or we may say that with /t/, absence of that component has 1-phoneme length so as not to exclude /a/ (which contains the component).

^18a Rather than include it in one of the vowel phonemes, which would confuse all the general statements about the distribution of vowel phonemes and their allophones.

¹⁹ In varying measure, this is the case in English (within a morpheme), Swahili, and Fanti.

²⁰ E.g. Fanti. See W. E. Welmers and Z. S. Harris, The Phonemes of Fanti, JAOS 62.319 (1942).

²¹ Y. R. Chao, op.cit. in fn. 4.

²² Leonard Bloomfield, The Structure of Learned Words, A Commemorative Volume Issued by the Institute for Research in English Teaching 17–23 (Tokyo, 1933).

^22a We have seen that the 2- and 3-place components of §5.2 and the fixed-sequence components of §5.1 differ in effect, in that the former describe limitations of distribution and the latter describe contours. It is of interest to notice wherein these two types of long components differ structurally and wherein they are similar.

They are similar in that they are all expressions for limitations of distribution of different segments. In the case of the pitch contours, we begin with allophonic segments that contain pitch and stress features in them. We notice that there are limitations upon the distribution of these segments. For instance, after a sequence of segments in which each loud-stressed segment is higher-pitched than the preceding, we never get a low-pitched segment: after Is your brother? we never get a low going, but only a going which is pitched even higher than brother. And in Is your brother going? we do not get a low pitched ing. We express this limitation of distribution by saying that all the segments of the utterance contain a particular component in common, and that this component has various phonetic values at various parts of its stretch: low pitch on the first low-stressed vowel, higher pitch on the next, etc. Exactly this is what we do with the 2- and 3-place components : We notice that after /s/ we never have /b/, but only /p/. We express this by saying that both successive segments have a particular component in common, and that this component has fortis value throughout its stretch.

The differences between the two types of long components are four. First, the phonetic values of the contour components are usually all pitch and stress features, which we are accustomed to consider a thing apart, while the phonetic values of the other components may seem to us to be arbitrarily extracted from the rest of the segment, as when we distinguish the closure of /b/ from its lip position.

Second, since the contour components are often constituents of simultaneous morphemes (e.g. the question intonation), we often cannot obtain the phonemes of the segmental morphemes (e.g. your or brother, without regard to intonation) until after the contour components have been extracted. Therefore we usually extract the contour components while working on sequences of allophonic segments, whereas we extract the other long components by working on sequences of phonemes.

Third, whereas the long components usually extend over a definite small number of phonemes, the contour components usually extend over a variable (and much larger) number—as many as there may be in a linguistic form or utterance of a particular type.

Fourth, we usually have many more positional variants of a contour component than of a 2- or 3-place component. The 1221130 of I don't know where he's going. (§5.1) and the 2230 of Bud Clark fumbled. are positional variants of the 230 in He told him. The phonemic component environment, which determines the number of 1's and the number and place of 2's in all these variants, is the simultaneous sequence of stress contours. The 2- or 3-place components usually have fewer though more complicated positional variants, as when the Swahili n component indicates tongue retarding with labial components, but velar occlusion with h (n + p = t, n + b = d, n + h = k).

²³ If only the first of these were true, we should have morphophonemic writing. We may permit partial overlapping among our components, i.e. the same sound feature may be represented in different environments by different components, but that is no bar to phonemic writing.

²⁴ This will in general happen only in cases of automatic morphophonemic alternation.

²⁵ See now Henry Hoenigswald, Internal Reconstruction, Studies in Linguistics 1944.

²⁶ Various other facts about the phonetic structure also transpire from a component analysis. One can tell, by looking at the combinations of components representing the phonemes, which phonemes ever occur next to each other and which never do (i.e. whether they have a long component in common), which phonemes replace each other in complementary environments (i.e. whether all their one-length components are identical), which phonemes have the smallest number of different phonemes next to them (i.e. the ones that contain the largest number of long components).